contech
contech copied to clipboard
The Contech analysis framework provides the means for generating and analyzing task graphs that enable computer architects and programmers to gain a deeper understanding of parallel programs.
https://github.com/bprail/contech/blob/b2cfe502a8dd21eb048bc96c961d95d3e497485f/backend/Heltech/heltech.cpp#L48 return in all paths return failure on races detected
how to deal with bbTime: bbDelta.cpp:89: virtual void BBDeltaBackend::updateBackend(contech::Task*): Assertion `currTime >= prevTime' failed.
If there are no actions of the expected type, then the collection initializes poorly and may not properly detect begin() == end(). Add an isEmpty() as well as better initialization...
Test: 1) Pure 1MB writes, initially artificial sized, can either pad out buffers or copy chunks in. 2) Async IO What is required to achieve 1+ GB/s?
The following "basic block" causes Contech to crash. ``` ; :37: ; preds = %33 br label %39 ; No predecessors! call void @llvm.trap() unreachable ```
What alias analysis support does LLVM provide? Could it support RAC further? Or could this be ruled out?
Email reported issue 1. I changed my makefile to run contech wrapper (contech_wrapper_par.py). However, the long timing cost problem has not been solved yet. I am using the ccv library,...
Change the tail duplication so that it only attempts when possible benefit. Could that require additional analysis?
Can the RAC approaches be integrated into a single pass on each block? Inter-/Intra-/Global detection. Loops are still separate.